Comparison of mathematics achievement of students with random forest, multi-layer based sensors and radial function neural network methods in terms of forecasting ability: Turkey example

In this dissertation study, in order to determine the factors affecting the Mathematics literacy of students participating in PISA 2015 test in Turkey, the methods of Random Forest from Decision Trees and Multilayer Perceptron Artificial Neural Networks (MLPANN) and Radial Basis Function Artificial Neural Networks (RBFANN) from Artificial Neural Networks (ANN) were implemented according to PISA 2015 data. The information, which was collected from a total of 4422 students, 2165 (49%)of whom were boys and 2257 (51%) of whom were girls having participated in PISA 2015 test, was used. The scores, which the students, having gone in for PISA 2015 test, got from mathematics test were included in the model as dependent variables; and 25 variables, which were thought to have connection with dependent variables institutionally, were included in the model as independent variables. In the study, three methods were compared in terms of predictability; and when the model was tested, it was observed that Random Forest method estimated the model with fewer errors in terms of a number of performance indicators compared to the other methods. Therefore, it is possible to say that Random Forest is more consistent in terms of performance criteria; and its estimating ability is a little higher compared to MLPANN and RBFANN. Moreover, when MLPANN and RBFANN methods are compared, it is concluded that MLPANN exhibited a better performance compared to RBFANN; on the other hand, when MLPANN and RF were compared, it was found out that RF was more successful.